an Abstraction for Large-Scale Computation

(1)

Experiences with MapReduce,

an Abstraction for Large-Scale Computation

Jeff Dean

Google, Inc.

(2)

Outline

•

Overview of our computing environment

•

^MapReduce

– overview, examples

– implementation details

– usage stats

•

Implications for parallel program development

(3)

Problem: lots of data

•

Example: 20+ billion web pages x 20KB = 400+ terabytes

•

One computer can read 30-35 MB/sec from disk – ~four months to read the web

•

~1,000 hard drives just to store the web

•

Even more to do something with the data

(4)

Solution: spread the work over many machines

•

Good news: same problem with 1000 machines, < 3 hours

•

Bad news: programming work – communication and coordination

– recovering from machine failure

– status reporting

– ^debugging – optimization

– ^locality

•

Bad news II: repeat for every problem you want to solve

(5)

Computing Clusters

•

Many racks of computers, thousands of machines per cluster

•

Limited bisection bandwidth between racks

(6)

Machines

•

^{2 CPUs}

– Typically hyperthreaded or dual-core

– Future machines will have more cores

•

1-6 locally-attached disks – 200GB to ~2 TB of disk

•

4GB-16GB of RAM

•

Typical machine runs:

– Google File System (GFS) chunkserver

– Scheduler daemon for starting user tasks

– One or many user tasks

(7)

Implications of our Computing Environment

Single-thread performance doesn’t matter

• We have large problems and total throughput/$ more important than peak performance

Stuff Breaks

• If you have one server, it may stay up three years (1,000 days)

• If you have 10,000 servers, expect to lose ten a day

“Ultra-reliable” hardware doesn’t really help

• At large scales, super-fancy reliable hardware still fails, albeit less often

– software still needs to be fault-tolerant

– commodity machines without fancy hardware give better perf/$

How can we make it easy to write distributed programs?

(8)

MapReduce

•

A simple programming model that applies to many large-scale computing problems

•

Hide messy details in MapReduce runtime library:

– automatic parallelization

– load balancing

– network and disk transfer optimization

– handling of machine failures

– ^robustness

– improvements to core library benefit all users of library!

(9)

Typical problem solved by MapReduce

• Read a lot of data

• Map: extract something you care about from each record

• Shuffle and Sort

• Reduce: aggregate, summarize, filter, or transform

• Write the results

Outline stays the same,

map and reduce change to fit the problem

(10)

More specifically…

• Programmer specifies two primary methods:

– map(k, v) <k', v'>*

– reduce(k', <v'>*) <k', v'>*

• All v' with same k' are reduced together, in order.

• Usually also specify:

–

partition(k’, total partitions) -> partition for k’

• often a simple hash of the key

• allows reduce operations for different k’ to be parallelized

(11)

Example: Word Frequencies in Web Pages

A typical exercise for a new engineer in his or her first week

•

Input is files with one document per record

•

Specify a map function that takes a key/value pair key = document URL

value = document contents

•

Output of map function is (potentially many) key/value pairs.

In our case, output (word, “1”) once per word in the document

“document1”, “to be or not to be”

“to”, “1”

“be”, “1”

“or”, “1”

(12)

Example continued: word frequencies in web pages

•

MapReduce library gathers together all pairs with the same key (shuffle/sort)

•

The reduce function combines the values for a key In our case, compute the sum

•

Output of reduce (usually 0 or 1 value) paired with key and

saved _{“be”, “2”}

“not”, “1”

“or”, “1”

“to”, “2”

key = “or”

values = “1”

“1”

key = “be”

values = “1”, “1”

“2”

key = “to”

values = “1”, “1”

“2”

key = “not”

values = “1”

“1”

(13)

Example: Pseudo-code

Map(String input_key, String input_value):

// input_key: document name

// input_value: document contents for each word w in input_values:

EmitIntermediate(w, "1");

Reduce(String key, Iterator intermediate_values):

// key: a word, same for input and output // intermediate_values: a list of counts int result = 0;

for each v in intermediate_values:

result += ParseInt(v);

Emit(AsString(result));

Total 80 lines of C++ code including comments, main()

(14)

Widely applicable at Google

– Implemented as a C++ library linked to user programs

– Can read and write many different data types

Example uses:

web access log stats web link-graph reversal inverted index construction statistical machine translation

… distributed grep

distributed sort

term-vector per host document clustering machine learning ...

(15)

Queries containing “eclipse” Queries containing “world series”

Example: Query Frequency Over Time

Queries containing “summer olympics”

Queries containing “Opteron”

Queries containing “full moon”

Queries containing “watermelon”

(16)

Example: Generating Language Model Statistics

•

Used in our statistical machine translation system

– need to count # of times every 5-word sequence occurs in large corpus of documents (and keep all those where count >= 4)

•

Easy with MapReduce:

– map: extract 5-word sequences => count from document

– reduce: combine counts, and keep if count large enough

(17)

Example: Joining with Other Data

•

Example: generate per-doc summary, but include per-host information (e.g. # of pages on host, important terms on host)

– per-host information might be in per-process data structure, or

might involve RPC to a set of machines containing data for all sites

•

map: extract host name from URL, lookup per-host info, combine with per-doc data and emit

•

reduce: identity function (just emit key/value directly)

(18)

MapReduce Programs in Google’s Source Tree

0 1000 2000 3000 4000 5000 6000

Jan-03 Jul-03 Jan-04 Jul-04 Jan-05 Jul-05 Jan-06 Jul-06

(19)

New MapReduce Programs Per Month

0 100 200 300 400 500 600

Jan-03 Jul-03 Jan-04 Jul-04 Jan-05 Jul-05 Jan-06 Jul-06

Summer intern effect

(20)

MapReduce: Scheduling

• One master, many workers

– Input data split into M map tasks (typically 64 MB in size) – Reduce phase partitioned into R reduce tasks

– Tasks are assigned to workers dynamically – Often: M=200,000; R=4,000; workers=2,000

• Master assigns each map task to a free worker – Considers locality of data to worker when assigning task – Worker reads task input (often from local disk!)

– Worker produces R local files containing intermediate k/v pairs

• Master assigns each reduce task to a free worker – Worker reads intermediate k/v pairs from map workers

– Worker sorts & applies user’s Reduce op to produce the output

(21)

Parallel MapReduce

(22)

Task Granularity and Pipelining

•

Fine granularity tasks: many more map tasks than machines – Minimizes time for fault recovery

– Can pipeline shuffling with map execution

– Better dynamic load balancing

•

Often use 200,000 map/5000 reduce tasks w/ 2000 machines

(23)

MapReduce status: MR_Indexer-beta6-large-2003_10_28_00_03

(24)

(25)

(26)

(27)

(28)

(29)

(30)

(31)

(32)

(33)

(34)

Fault tolerance: Handled via re-execution

On worker failure:

•

Detect failure via periodic heartbeats

•

Re-execute completed and in-progress map tasks

•

Re-execute in progress reduce tasks

•

Task completion committed through master On master failure:

•

State is checkpointed to GFS: new master recovers &

continues

Very Robust: lost 1600 of 1800 machines once, but finished fine

(35)

Refinement: Backup Tasks

•

Slow workers significantly lengthen completion time – Other jobs consuming resources on machine

– Bad disks with soft errors transfer data very slowly

– Weird things: processor caches disabled (!!)

•

Solution: Near end of phase, spawn backup copies of tasks – Whichever one finishes first "wins"

•

Effect: Dramatically shortens job completion time

(36)

Refinement: Locality Optimization

Master scheduling policy:

•

Asks GFS for locations of replicas of input file blocks

•

Map tasks typically split into 64MB (== GFS block size)

•

Map tasks scheduled so GFS input block replica are on same machine or same rack

Effect: Thousands of machines read input at local disk speed

•

Without this, rack switches limit read rate

(37)

Refinement: Skipping Bad Records

Map/Reduce functions sometimes fail for particular inputs

• Best solution is to debug & fix, but not always possible

On seg fault:

– Send UDP packet to master from signal handler

– Include sequence number of record being processed

If master sees K failures for same record (typically K set to 2 or 3) :

• Next worker is told to skip the record

Effect: Can work around bugs in third-party libraries

(38)

Other Refinements

•

Optional secondary keys for ordering

•

Compression of intermediate data

•

Combiner: useful for saving network bandwidth

•

Local execution for debugging/testing

•

User-defined counters

(39)

Using 1,800 machines:

• MR_Grep scanned 1 terabyte in 100 seconds

• MR_Sort sorted 1 terabyte of 100 byte records in 14 minutes

Rewrote Google's production indexing system

• a sequence of 7, 10, 14, 17, 21, 24 MapReductions

• ^simpler

• more robust

• ^faster

• more scalable

Performance Results & Experience

(40)

MR_Sort

(41)

Usage Statistics Over Time

426 55 3,351 1.2 157 193 758 3,288 217 634 Aug, ‘04 29,423

2345 147 3,836 5.0 268 2,970 6,743 52,254 2,002 874 Mar, ‘06 171,834

3,097 Average map tasks per job

411 Unique map/reduce combinations

144 Average reduce tasks per job

1.9 Average worker deaths per job

232 Average worker machines

941 Output data written (TB)

2,756 Intermediate data (TB)

12,571 Input data read (TB)

981 Machine years used

934 Average completion time (secs)

Mar, ‘05 72,229 Number of jobs

(42)

Implications for Multi-core Processors

• Multi-core processors require parallelism, but many programmers are uncomfortable writing parallel programs

• MapReduce provides an easy-to-understand programming model for a very diverse set of computing problems

– users don’t need to be parallel programming experts

– system automatically adapts to number of cores & machines available

• Optimizations useful even in single machine, multi-core environment

– locality, load balancing, status monitoring, robustness, …

(43)

Conclusion

• MapReduce has proven to be a remarkably-useful abstraction

• Greatly simplifies large-scale computations at Google

• Fun to use: focus on problem, let library deal with messy details

– Many thousands of parallel programs written by hundreds of different programmers in last few years

– Many had no prior parallel or distributed programming experience

Further info:

MapReduce: Simplified Data Processing on Large Clusters, Jeffrey Dean and Sanjay Ghemawat, OSDI’04

http://labs.google.com/papers/mapreduce.html (or search Google for [MapReduce])